Re: [syzbot] general protection fault in list_lru_add

[Muchun Song <songmuchun@bytedance.com>]

[-- Attachment #1: Type: text/plain, Size: 2826 bytes --]

On Fri, Mar 25, 2022 at 12:18 AM Dmitry Vyukov <dvyukov@google.com> wrote:
>
> On Thu, 24 Mar 2022 at 17:13, Muchun Song <songmuchun@bytedance.com> wrote:
> >
> > On Thu, Mar 24, 2022 at 4:50 PM Dmitry Vyukov <dvyukov@google.com> wrote:
> > >
> > > On Thu, 24 Mar 2022 at 09:44, Muchun Song <songmuchun@bytedance.com> wrote:
> > > >
> > > > On Thu, Mar 24, 2022 at 11:05 AM Linus Torvalds
> > > > <torvalds@linux-foundation.org> wrote:
> > > > >
> > > > > On Wed, Mar 23, 2022 at 7:19 PM Muchun Song <songmuchun@bytedance.com> wrote:
> > > > > >
> > > > > > After this commit, the rules of dentry allocations changed.
> > > > > > The dentry should be allocated by kmem_cache_alloc_lru()
> > > > >
> > > > > Yeah, I looked at that, but I can't find any way there could be other
> > > > > allocations - not only are there strict rules how to initialize
> > > > > everything, but the dentries are free'd using
> > > > >
> > > > >         kmem_cache_free(dentry_cache, dentry);
> > > > >
> > > > > and as a result if they were allocated any other way I would expect
> > > > > things would go south very quickly.
> > > > >
> > > > > The only other thing I could come up with is some breakage in the
> > > > > superblock lifetime so that &dentry->d_sb->s_dentry_lru would have
> > > > > problems, but again, this is *such* core code and not some unusual
> > > > > path, that I would be very very surprised if it wouldn't have
> > > > > triggered other issues long long ago.
> > > > >
> > > > > That's why I'd be more inclined to worry about the list_lru code being
> > > > > somehow broken.
> > > > >
> > > >
> > > > I also have the same concern.  I have been trying for a few hours to
> > > > reproduce this issue, but it didn't oops on my test machine.  And I'll
> > > > continue reproducing this.
> > >
> > > syzbot triggered it 222 times in a day, so it's most likely real:
> > > https://syzkaller.appspot.com/bug?extid=f8c45ccc7d5d45fc5965
> > >
> > > There are 2 reproducers, but they look completely different. May be a race.
> > > You may also try to use syzbot's patch testing feature to get some
> > > additional debug info.
> >
> > Do you know how to tell the syzbot to test the following patch?
> > I found some infos from github, it says "#syz test:", is it like the following?
> > Thanks.
> >
> > #syz test: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
> > master
>
> Yes, this is correct. You can now see the request listed here:
> https://syzkaller.appspot.com/bug?extid=f8c45ccc7d5d45fc5965
>

Cool!.

> but the patch was truncated (probably you email client messed
> whitespaces). In such case it's more reliable to attach the patch as
> text file.

Thanks for your reminder.

#syz test: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
5abc1e37afa0335c52608d640fd30910b2eeda21

[-- Attachment #2: test.patch --]
[-- Type: application/octet-stream, Size: 822 bytes --]

diff --git a/mm/list_lru.c b/mm/list_lru.c
index c669d87001a6..ddb2ee627d32 100644
--- a/mm/list_lru.c
+++ b/mm/list_lru.c
@@ -67,6 +67,7 @@ list_lru_from_kmem(struct list_lru *lru, int nid, void *ptr,
 	struct list_lru_node *nlru = &lru->node[nid];
 	struct list_lru_one *l = &nlru->lru;
 	struct mem_cgroup *memcg = NULL;
+	int kmemcg_id;
 
 	if (!list_lru_memcg_aware(lru))
 		goto out;
@@ -75,7 +76,13 @@ list_lru_from_kmem(struct list_lru *lru, int nid, void *ptr,
 	if (!memcg)
 		goto out;
 
-	l = list_lru_from_memcg_idx(lru, nid, memcg_kmem_id(memcg));
+	kmemcg_id = memcg_kmem_id(memcg);
+	l = list_lru_from_memcg_idx(lru, nid, kmemcg_id);
+	if (!l) {
+		pr_info("BUG: the memcg(%px)->objcg(%px), kmemcg_id: %d\n",
+			memcg, memcg->objcg, kmemcg_id);
+		BUG();
+	}
 out:
 	if (memcg_ptr)
 		*memcg_ptr = memcg;